The State of the Semantic Layer: 2025 in Review
One of the key differences between https://aicurated.com/ an ontology and a plain semantic model is the reasoning and inference dimension. A semantic layer sits between raw data and the people or systems that consume it. They need governed business meaning, reliable relationships, connected knowledge, and context about what matters right now.
We shipped one of the first enterprise semantic layers for large-scale production deployments when most modern data tools didn’t exist. Founded in 1989, Strategy, then MicroStrategy, was among the first companies to architect the layer between raw data and business meaning. A semantic layer should be managed with the same rigor as software and data systems, not edited ad hoc inside downstream dashboards or reports.
Large Language Models (LLMs) are powerful but unreliable when querying enterprise data—our research shows that they can be wrong 80% of the time when applied to structured analytics. Moderated by Jim Halcomb, Global Head of Research & Development at EDM Council, the discussion explored how semantic layers serve as the cornerstone of modern enterprise data strategies. Organizations that implement a governed semantic layer reduce time spent on reconciliation and shift focus back to analysis and decision-making. Metrics are defined once and enforced at query time, so every tool, team, and AI system returns the same answer. Both technical and non-technical users can define and extend models, while AI systems consume governed definitions Recommendation on AI ethics directly for reliable analytics. Mosaic translates raw data structures into the language your organization actually uses, ensuring revenue means the same thing in every dashboard, AI query, and report.
Semantic model
An agent answering a question about a customer’s renewal risk retrieves recent support tickets, open opportunities, and usage trends. An agent combining product usage data with support ticket data pulls from two systems that define the customer identifier differently. A context graph adds situational relevance so AI systems understand what matters in the moment. That structure gives systems a governed model of business meaning. An ontology can support AI Act rules, classifications, and inference-style reasoning.
Why a governed semantic layer is critical for enterprise AI
A knowledge graph connects entities and relationships in a form that supports reasoning and traversal across domains. All four help systems understand and work with data, but they do it in different ways and at different levels. Beyond her client work, she serves as California State Ambassador for Women in AI, a global nonprofit advancing gender diversity in artificial intelligence, where she supports mentorship and advocacy for women building careers in the field. It’s increasingly about building and maintaining the Limitations of Semantic Layers for Enterprise AI Context context that lets a machine interpret data correctly without a human in the loop to catch the ambiguity. The job used to be building reports and charts for humans to interpret with their own judgment layered on top. It’s a deliberate semantic layer — the shared, machine-readable set of business definitions that sits between your raw, clean data and whatever model or agent is querying it.
The Path Every Question Follows Through the Contextual Semantic Layer
- Dbt Semantic Layer is a separate capability within dbt that adds business metric definitions on top of those transformed tables.
- The undefined, inconsistent, undocumented metric definitions are.
- For organizations running four or more BI tools, deploying AI agents at scale, or building embedded analytics products, the universal semantic layer is the correct long-term architecture.
- The universal semantic layer (also called a headless semantic layer) is a dedicated, platform-agnostic layer that sits between your data warehouse and all your downstream consumers — BI tools, AI agents, embedded analytics, custom applications.
- The fix is to govern and certify your core metrics in the Power BI semantic model before enabling Copilot organization-wide.
- Power BI Copilot queries your semantic model, not your raw data.
Logic is defined once, as a business concept, and applied consistently across tools, applications, and AI systems. Metric definitions live in a dedicated semantic layer that every downstream system queries through. AI systems querying those tables still lack access to the business meaning behind the data.
Semantic Layer Architecture: Components, Design Patterns, and AI Integration – Databricks
Semantic Layer Architecture: Components, Design Patterns, and AI Integration.
Posted: Fri, 20 Mar 2026 23:02:12 GMT source